We describe events that led to successful testing of melphalan, one of the nitrogen mustard compounds, in children with newly diagnosed, poor-risk rhabdomyosarcoma (RMS). Preclinical studies with xenografts of human RMS, growing in the flanks of immune-deprived mice, had indicated superior oncolytic activity by melphalan compared with other agents commonly used to treat this tumor. However, in a conventional phase II trial, melphalan failed to produce partial responses in 12 of 13 heavily pretreated patients with recurrent tumors. Subsequent comparison of the drug's pharmacokinetics in mice and patients indicated that its poor clinical performance was not the result of interspecies differences in drug disposition. Therefore, we elected to retest melphalan in untreated patients, before they were enrolled in a phase III study. Of 13 children who received the drug for 6 weeks, ten had partial responses, confirming the significant antitumor activity seen in the xenograft system. These findings illustrate the inherent limitations of phase II drug trials in previously treated patients and suggest a useful paradigm for the development of antineoplastic drugs.